Winning the Unwinnable — LessWrong